Spaced repetition system (or SRS) is a way to memorize “items” of knowledge (i.e. vocabulary), by making the student rack his or her brain trying to remember them at increasing intervals. You can find a bit more detailed description about it on Wikipedia. If you tried Anki or the long-term study in zkanji you probably know what I’m talking about. I don’t intend to write about the way SRS works and would avoid any scientific stuff. What follows has probably been said already (I don’t know, I haven’t read what others wrote), but this is a blog, so I just post what I think.
Let me make it clear that I think SRS is a great way to remember things. It has received a lot of criticism (probably, I have only seen a few, but I don’t roam the internet trying to find out what everyone thinks), and some of those partially concur with what I think as well. People who criticize it probably have their reasons to do so, and I’m not going to debunk them, nor am I able to. I just wanted to write about one correct way of using SRS.
This might sound like the opposite of what I wrote so far, but I think SRS is not the best method when you want to learn something new. Although it might be a way to study, but it is rather a way to prevent forgetting what you studied before. If you are like me and don’t use Japanese in your daily life, you ought to forget even some pretty common vocabulary. Preventing this is one purpose of SRS. I have met the opinion that using this method is a waste of time, and that many people learned a language without using spaced repetition. I agree that SRS is not for everyone, and that your experience might tell you to avoid it. There are people who learned a language without every touching textbooks, and textbooks might have slowed them down even. It depends on your environment and skills. If you have the opportunity to communicate a lot in your target language or you are surrounded with that language all the time, the correct way of using SRS might be to not use it at all. It can still be useful, so it is up to every person whether they need it or not. It is also possible that SRS wouldn’t help you because it doesn’t suit your learning style. For example books for learning kanji never helped me, even though a lot of students found them helpful.
I don’t want to tell anyone what they should do, but I have a few ideas regarding the use of spaced repetition systems. I think they are only effective if you combine them with some other study method. For example reading is both great entertainment and an excellent way to study. Even the best writers tend to use the same words and expressions over and over again, so if you read several books from the same author, you can learn without SRS. You could even take reading as a badly tuned SRS, though it doesn’t force you to remember as much as a program used for testing you. Especially if you are lazy and never take the time to learn what you have read. :p Reading Japanese books is “a bit” difficult due to kanji, but fortunately it is not the only way. You could pick any kind of entertainment, collect words you heard or read and study those. It is not a good idea to just add new words to the list (or deck) of an SRS program, because spaced repetition is not effective that way. My opinion is that if you make sure that what you learn with SRS can be associated with something else that you enjoy or use (entertainment, textbooks etc.), and that you try to be exposed to the Japanese language (or any other subject) as much as possible, it is easier to remember things with, than without SRS.
I could write a lot more about the subject, but I’m not confident in my authoritativeness. Just try it before you form an opinion.
As a final word I wanted to tell this to those who complain that the intervals given to words in the zkanji long-term study are too long. Repeating the words with the normal study groups, before adding them to the long-term study list helps a lot. At least if you try to learn brand new vocabulary.
The good news is, the new long-term study list is done. But it’s not tested at all. I could do as a good software company and release it for everyone, expecting error reports to come in so I don’t have to hunt for bugs myself, but being a perfectionist I don’t want to do it like that. I will only release the next version when I have implemented everything I planned.
A quick listing of what’s missing:
- Automatic inclusion of JLPT words for different levels.
Needed changes to the “handwritten” kanji recognition window inside the test: Not showing kanji popup, which might help too much. Inclusion of the different kurikaeshi symbols, so all characters can be typed for written form of words. Inclusion of kana. (This has nothing to do with testing because kana can still be typed with the keyboard, but I can already see some people complaining about it.) Only recognize kanji when the stroke order was perfect. (In the global recognition window you can reverse any 2 strokes.)
- UPDATE (Dec. 22, 2010): Decided not to make this yet.
Updating the kanji database with the new jlpt N level. (This will also require some changes to the file format because I want to keep the old kyuu level as well)
- UPDATE (Dec. 22, 2010): Instead of this, a better menu structure was made.
Customizable toolbar on the single window interface, because finding some functions are so difficult in the menu.
Having said that, you can still read the title for today’s entry. The test version will be announced here and I will put a link on the zkanji site that anyone can download.
Kanji Information window
I have made changes to how the kanji information window (that appears when you double-click a kanji) looks like, because it was the same cluttered interface since day 1. From the next version (included in tomorrow’s test) the kanji information window will only show those reference numbers that the user selects either in the settings or in the popup-menu of the info window. Their order can be changed as well. I plan to add coloring to different numbers, but I thought that it’s not so important right now.
Kanji Information window with all reference numbers displayed.
Kanji Information window with the radical, a few reference numbers and their order changed.
Kanji Information window with only the Stroke Order Diagram displayed.
As I promised I will report on the progress of the rewritten long-term study list. It’s not finished yet but it’s progressing well and I expect it done within a few days. There is not much I can illustrate with screenshots, as most of the changes affected parts of the program that prevented me from rewriting the study list before.
Managing long-term study list items:
After adding items and pressing the “Long-term study” button, the student is presented with all the words added to the list. There are currently 2 views on the window for viewing the new words queued for testing and for browsing the already tested items and their study data. (Statistics not done yet.) This window also has an input field for changing the meaning of words in tests, so if you missed something in the import process, you can still make those changes.
In cases when a word doesn’t have all items included for testing out of the three possibilities (written, kana form and meaning), the user can invoke the import window from a popup menu. Test priorities can be changed from that menu too. I didn’t want to include this option at first, because I personally like controlling the exact order of new words in the test, but I had to change my mind for practical reasons. The word listing can be reordered by several columns and filtered (eg. by kana reading), so it wouldn’t be easy to change their test order unless I added another view where the items are only presented in their order of inclusion, but that would have resulted in bad user experience. (Imagine having to decide what each view does when they all look almost the same.)
There are two input fields here. The top one is for entering meaning, the bottom one is for entering kanji/kana during the test. These fields will only be present if the student wants them to (it depends on the settings), but they won’t affect the outcome. The buttons for selecting the result will still be presented. Notice that there are 4 buttons for this. Once again I have changed my mind and dismissed the idea of only using the 2 buttons from the first long-term study test. The 2 main buttons are still there in the usual position, and they will work similarly to the original version. The first one makes the word get repeated “soon” in the same session, lowering its score, while the second one is for accepting the answer and moving on.
On the far right there are two new buttons for better control. More often than not I knew the answer to a specific question, but felt that I needed to review them too soon. That’s why there is one button for doing that without affecting the word’s score. The result is very similar to the “Can’t recall” button in that the word will be repeated soon, probably in the same session. There is also a fourth button for cases when the word is so familiar it got boring.
There is another change compared to the first version of the test. The student can now input the written form of the word including the kanji, using the written kanji recognition. It still can’t recognize kana input drawn with the mouse (you will have to type that), but with this feature practicing writing kanji is a possibility too.
Problems with automatically selecting new items to test:
I’m currently working on composing the list of words to be included in a testing session. Those who’ve never tried designing such a system probably can’t understand why does it take so much time and why is it so difficult. This task was much easier in the old long-term study, because zkanji picked the words that were due at a given day in the order of their date to be tested, and the new words were included at the front of the test. It was simple and straightforward. This have changed with item priorities and dynamically changing lists. The test can be stopped any time and the result must be saved immediately, and zkanji should remember failed cards to be able to present them soon again, which makes things difficult. The biggest problem (at the moment) is not this though. The one thing giving me headache right now (apart from not sleeping enough) is that I want to space out the items for the same word to different study sessions, and if possible different days. If two questions about the same word appear too close to each other the student might still remember the answer from short-term memory.
I don’t want to get into details more than this because the algorithm for deciding the order of tested items is not written yet. See you next time after I have (hopefully) solved all the problems!
In the previous post I have outlined what I expect from the rewritten test. Since then I spent some time thinking about how to do all that, and although writing the code itself doesn’t look too complicated from here (before actually starting it), it’s always the planning that takes up most of the time. In the previous years I just skipped the planning phase and went right to the coding, this is why I’m now in the not so pleasant situation where I have to throw out an important part of zkanji (which is very difficult), and make it anew. Going ahead without planning worked before as I knew exactly what I wanted, but as it turned out, there are problems which require some experience in failing to solve them.
This time I have it all planned out. It’s only in my head at the moment, that’s why I’m going to share some of it here. “Just do it already!”. Ah yes, sorry. First let me show you what is done and how it works: (click on the thumbnails to view the original images)
Selecting words for the test:
The first step is the same as before. The user collects kanji in a kanji group for studying, and hand-picks the words for each of them to include in the long-term study list. The picked words have red and yellow colored background, depending on the selected reading on the far right. (yellow = words that have the kanji with the selected reading, red = words with the same kanji but different reading)
There is no visible change here at the moment, but I plan to add an important improvement to the selection process. A search field would make picking the example words faster, and there is enough space for it next to the M and P buttons. (Move your mouse over the buttons to see a tooltip explaining what they do.)
In the finished version this kind of hand-picking might become obsolete as there will be automatic word selection for studying, but if the student wants to control the selection process, this method will still be available.
Importing items into the study list:
In previous versions when the user pressed the button with the red arrow pointing down next to the Long-term study button, there was no reaction from the program. The words were added to the list, but the lack of any indication of this confused many users.
Not just that, but the user had no control over the items to include. With each word, two items (or cards, if you prefer) were added. One for the written form of a word (with kanji in it), and one for the reading (kana only). Learning both of these can be important for a beginner, but more often than not I have seen items I knew from the beginning. I just couldn’t exclude them from the test. Apart from the kanji and kana, there is now an option to add an item for testing the meaning of a word too.
On this form the user can check or uncheck items to be included or left out from testing. Words with all three items on the study list won’t be shown, and the check boxes for items that are already on the list will be grayed out.
There is an input field at the bottom of the form, where the user can change the meaning to be tested. Learning shorter definitions are easier than longer ones, and because the dictionary includes many synonyms in the definition of words, it makes sense to just remove most of them from the meaning. When everything is done and the OK button is pressed, the long-term study list will contain the items that were selected.
Managing items in the long-term study list: (This part is not done yet, so I can only write about my plans.)
After adding words to the study list and pressing the Long-term study button beside the kanji groups, a new window will open where the student can check previous study statistics and modify the list of items to be tested. I don’t have plans about the statistics yet, so let’s talk about the latter. The list of items can be short but also very long, so managing it might not be easy. I have tried Anki, and saw that in that program each card can have a priority assigned to it. This priority could later determine the order of the questions, and there was also an option to make the program pick the cards randomly (but still taking priority into consideration).
Unfortunately I don’t think that anything like this can be used in zkanji. If the program automatically selects the items to be tested, those items will be somewhat random anyway. The student might want to, say, test himself or herself on N5 items first, then N4 etc. before getting to N2, or probably take the more popular words first. Therefore there might be an option to do this, but it’s not easy to say anything before I have a list for the words required on the N levels. (I hope I can find someone who is willing to give me one for free…)
I plan to make studying very similar to studying in Anki but not the same. The 5 buttons in Anki are overwhelming, so the long-term study list’s test window will have 3 instead. In previous versions of zkanji only 2 buttons were on the test window which were just enough, but in those versions the test didn’t finish until all items were answered correctly once or incorrectly 3 times. There was no need for a button to test the same word “soon”.
Hopefully my next post will come after finishing the long-term study list, but if I can’t do it in a week, I will return to report on my progress. (I won’t write this much though, just post some screenshots.)
The long-term study list is a feature that I added in the first quarter of 2009. It was an attempt at the so-called spaced repetition learning technique for remembering kanji. I added it in a hurry because I was planning to take the then 2 kyuu of the JLPT (which was later renamed to N2 in 2010). I knew less than a 100 kanji at the time and had 10 months to memorize the required 1000 kanji to pass. The long-term study list was a success and I finished remembering around 1300 kanji 8 months later, and also passed 2 kyuu.
Actually this feature turned out to be just another way to test the student’s knowledge of words. The main difference between this test and the previous word tests that already existed in zkanji at the time is that the student should use it daily, and that the program automatically picks the words to include in the day’s test. Another difference is that the words in the long-term study list come from example words previously selected for kanji (by the student), and the daily test result influences when the words are picked again. You could say that so far this doesn’t sound like a way to study kanji at all, but it worked for me pretty well. Even though it only handles words, all of them have to contain one or more kanji as they come from kanji example words. Also because these words are first selected for kanji, many words that are asked at their first inclusion contain the same kanji.
After using the long-term study list for over a year I have found several flaws in its (that is my) design. First of all, adding 20 words to the list every day (so I can memorize the required number of words till the JLPT) made the automatically picked amount grow above 300 items on some days. Even with the average 150 items the test took me 2 hours to complete each day, not to mention that I had to finish the whole test on the same day or no result was recorded. So even though the study was successful, testing myself with the long-term study list have become a burden, and I have stopped doing it altogether.
To prevent forgetting everything I have learned so far and also to be able to learn new words I have decided to completely rewrite the long-term study list based on my past experience. The new test will still concentrate on words, but it should be able to automatically add new words to the test depending on the student’s aim. (Eg. taking the N3.) Other planned improvements include:
- ability to stop the test and still save the results, so the effort won’t be wasted
- giving several options for the student to decide when the tested word will be included next – this will not only give more control over the interval, but also free up time otherwise spent with the test, because some words are just too easy to remember to test them so often
- adding words to the test that don’t contain kanji – as many are required by the JLPT but are also useful in themselves anyway. There is no use for limiting the words that can be tested either.
- testing the meaning of words based on the written form (containing kanji) and reading (only kana) – this is not so useful to me, but was requested by a friend
What makes the whole rewriting business difficult is that the previous test results shouldn’t be lost. Unfortunately I found no way to keep the past statistics meaningful, but it would cause problems to just throw them away, so they will probably be converted to some garbage. Unless I find a good method.
I don’t want to compete with programs like Anki which can be used for studying so many different things, but I have to admit that I’ve always hated its user interface. Otherwise I might just make a way to export decks to Anki from zkanji (which could be a useful feature actually).